[CostModel][X86] Add bool anyof/allof reduction costs

On pre-AVX512 targets we can use MOVMSK to extract reduced boolean results. This is properly optimized, annoyingly AVX512 isn't and produces code that is almost as bad as the (unchanged) costs suggest......

Differential Revision: https://reviews.llvm.org/D60403

llvm-svn: 358574

1 files changed, 42 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index ca32ae093cd..b0a45fba17b 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -2489,6 +2489,48 @@ int X86TTIImpl::getArithmeticReductionCost(unsigned Opcode, Type *ValTy,
         return LT.first * Entry->Cost;
   }
 
+  static const CostTblEntry AVX2BoolReduction[] = {
+    { ISD::AND,  MVT::v16i16,  2 }, // vpmovmskb + cmp
+    { ISD::AND,  MVT::v32i8,   2 }, // vpmovmskb + cmp
+    { ISD::OR,   MVT::v16i16,  2 }, // vpmovmskb + cmp
+    { ISD::OR,   MVT::v32i8,   2 }, // vpmovmskb + cmp
+  };
+
+  static const CostTblEntry AVX1BoolReduction[] = {
+    { ISD::AND,  MVT::v4i64,   2 }, // vmovmskpd + cmp
+    { ISD::AND,  MVT::v8i32,   2 }, // vmovmskps + cmp
+    { ISD::AND,  MVT::v16i16,  4 }, // vextractf128 + vpand + vpmovmskb + cmp
+    { ISD::AND,  MVT::v32i8,   4 }, // vextractf128 + vpand + vpmovmskb + cmp
+    { ISD::OR,   MVT::v4i64,   2 }, // vmovmskpd + cmp
+    { ISD::OR,   MVT::v8i32,   2 }, // vmovmskps + cmp
+    { ISD::OR,   MVT::v16i16,  4 }, // vextractf128 + vpor + vpmovmskb + cmp
+    { ISD::OR,   MVT::v32i8,   4 }, // vextractf128 + vpor + vpmovmskb + cmp
+  };
+
+  static const CostTblEntry SSE2BoolReduction[] = {
+    { ISD::AND,  MVT::v2i64,   2 }, // movmskpd + cmp
+    { ISD::AND,  MVT::v4i32,   2 }, // movmskps + cmp
+    { ISD::AND,  MVT::v8i16,   2 }, // pmovmskb + cmp
+    { ISD::AND,  MVT::v16i8,   2 }, // pmovmskb + cmp
+    { ISD::OR,   MVT::v2i64,   2 }, // movmskpd + cmp
+    { ISD::OR,   MVT::v4i32,   2 }, // movmskps + cmp
+    { ISD::OR,   MVT::v8i16,   2 }, // pmovmskb + cmp
+    { ISD::OR,   MVT::v16i8,   2 }, // pmovmskb + cmp
+  };
+
+  // Handle bool allof/anyof patterns.
+  if (ValTy->getVectorElementType()->isIntegerTy(1)) {
+    if (ST->hasAVX2())
+      if (const auto *Entry = CostTableLookup(AVX2BoolReduction, ISD, MTy))
+        return LT.first * Entry->Cost;
+    if (ST->hasAVX())
+      if (const auto *Entry = CostTableLookup(AVX1BoolReduction, ISD, MTy))
+        return LT.first * Entry->Cost;
+    if (ST->hasSSE2())
+      if (const auto *Entry = CostTableLookup(SSE2BoolReduction, ISD, MTy))
+        return LT.first * Entry->Cost;
+  }
+
   return BaseT::getArithmeticReductionCost(Opcode, ValTy, IsPairwise);
 }